131 results about "Gradient estimation" patented technology

A novel method and apparatus for improving transmit antenna weight tracking using channel vector element to element correlations in a wireless communication system is disclosed. The present channel autocorrelation tracking technique utilizes the observation that tracking can be improved when a channel gain vector contains correlated elements. In a first embodiment of the autocorrelation tracking technique, the present invention extracts a coarse gradient estimate by utilizing a perturbation vector autocorrelation matrix estimate and a perturbation autocorrelation matrix to update TxAA weight vectors accordingly. In a second embodiment of the channel autocorrelation tracking technique, the present invention extracts a coarse gradient estimate by utilizing eigendecompositions, perturbation vector autocorrelation matrix estimates, and perturbation autocorrelation matrices to update TxAA weight vectors accordingly. In a third embodiment of the channel autocorrelation tracking technique, the present invention reduces the phase change that can occur at receivers.

A novel method and apparatus for applying overlaid perturbation vectors for gradient feedback transmit antenna array adaptation is disclosed. The method and apparatus of the present invention allows a communication system to reduce transmit power that is associated with dedicated pilot signals by overlaying perturbation vectors and measuring channel estimates and demodulation channel estimates during a measurement time interval that comprises a plurality of feedback time intervals. The present inventive method utilizes channel estimates that include the effects of previous perturbation vectors, subsequent feedback vectors and intermediate feedback decisions. The inventive method extracts a coarse gradient estimate by utilizing a continuous summation of overlaid weight vector perturbation vectors and updates the weighting vector accordingly.

The invention discloses a method for jointly estimating vehicle mass-road gradient under the consideration of influence of braking and turning. The method includes steps of 1, acquiring data, to be more specific, 11, acquiring vehicle driving state data by data acquisition devices, 12, combining the vehicle driving state data and vehicle intrinsic parameters with one another and computing relevant parameters required by models; 2, building relation models of vehicle mass, road gradient and the vehicle driving state data on the basis of vehicle dynamic models; 3, building least square mass estimation models and Kalman filter gradient estimation models on the basis of the relation models of the vehicle mass, the road gradient and the vehicle driving state data; 4, jointly estimating the vehicle mass and the road gradient by means of nested loop iteration; 5, eliminating the influence of braking and turning in state maintenance modes. The method has the advantages that the influence of braking and turning is considered in real vehicle driving procedures, and accordingly the vehicle mass and the road gradient can be dynamically jointly estimated in real time.

A method and system for determining an optical flow field between a pair of images is disclosed. Each of the pair of images is decomposed into image pyramids using a non-octave pyramid factor. The pair of decomposed images is transformed at a first pyramid scale to second derivative representations under an assumption that a brightness gradient of pixels in the pair of decomposed images is constant. Discrete-time derivatives of the second derivative image representations are estimated. An optical flow estimation process is applied to the discrete-time derivatives to produce a raw optical flow field. The raw optical flow field is scaled by the non-octave pyramid factor. The above-cited steps are repeated for the pair of images at another pyramid scale until all pyramid scales have been visited to produce a final optical flow field, wherein spatiotemporal gradient estimations are warped by a previous raw optical flow estimation.

A vehicle and vehicle system are provided with a controller that is configured to generate output indicative of a vehicle mass estimation. The vehicle mass estimation is based on a longitudinal acceleration and a wheel torque when at least one of the longitudinal acceleration, a vehicle speed and a yaw rate indicate an occurrence of a qualified event. The controller is further configured to generate output indicative of a dynamic road gradient estimation based on the vehicle speed, the wheel torque and the vehicle mass estimation.

A method for controlling a vehicle that includes a motor configured to provide a driving/braking force to the vehicle and a friction braking mechanism configured to provide a friction brake force to the vehicle includes a target calculation step of calculating a target torque of the motor in accordance with a displacement of an accelerator pedal, a gradient estimation step of estimating a gradient torque to cancel a disturbance due to a gradient of a road surface where the vehicle is travelling, a command calculation step of calculating a torque command value of the motor based on the gradient torque and the target torque, a control step of controlling a torque of the motor in accordance with the torque command value, and a stop control step.

The invention relates to a guided strategy search reinforcement learning algorithm. Firstly, a guided learning sample is selected, and then the selected sample is used to perform gradient estimation on an objective function, and parameters are updated according to a policy update principle until convergence. The invention greatly reduces the problem of lowering algorithm stability performance and convergence rate due to the use of important sampling technology by reconstructing the objective function. The invention defines a guided high-quality learning sample for reinforcement learning, through the use of the guided learning sample, the policy search can be performed more accurately, thereby avoiding a local optimum in a bad situation.

The invention discloses a bistatic SAR motion compensation method based on phase gradient autofocus improvement. The method mainly solves the problem that in the prior art, motion compensation cannot be carried out under the condition of large motion errors. The method comprises the complementation steps that 1) range migration correction, range compression and deramp processing are carried out on sub-aperture data; 2) non-empty variable phase error estimation is carried out on signals through a PGA method, rough compensation is carried out on the data through the error estimation result, and residual phase errors are obtained; 3) the residual phase errors are unfolded to be a second-order polynomial of the range, and a constant term coefficient and linear and quadratic term coefficients are obtained; 4) the minimum mean square gradient estimation of the three coefficients is obtained through a phase weighting PGA, and the sub-aperture phase errors are obtained through the gradient estimation result; 5) the phase errors of the sub-aperture data are spliced to obtain full-aperture phase errors; 6) motion compensation and bistatic SAR imaging are carried out through the full-aperture phase errors. The method can be used for processing the bistatic SAR data under the condition of the large motion errors.

The invention discloses a defense method and a defense device for a black box attack model of a speech recognition system. According to the defense method, simulated environment noise is added to an original audio; a voice input condition is simulated in a real scene; noise is randomly added to form a primary adversarial sample; the adversarial sample is optimized through a genetic algorithm andgradient estimation to obtain an accurate adversarial sample, and then the original audio file and the adversarial sample are mixed to serve as a training data set of adversarial training to train themodel, so that the recognition accuracy of the model to the adversarial sample is improved, and the robustness of the model to the adversarial attack is improved.

The invention discloses a multi-direction seismic energy gradient difference carbonate karst cave type reservoir identification method. The method comprises the following steps of: converting overlapped seismic data into a seismic energy data volume through Hilbert conversion; selecting two different directions according to the spatial characteristics of a carbonate karst cave type reservoir; performing gradient estimation by using a least square method; and calculating the gradient difference of different directions. The multi-direction seismic energy gradient difference carbonate karst cave type reservoir identification method has the advantages that a linear fitting method is stable, estimation radius can be randomly defined, the spatial position and shape of a carbonate karst cave can be accurately determined, the carbonate karst cave can be effectively identified, and the prediction accuracy and efficiency of the carbonate karst cave type reservoir can be remarkably improved.

The invention discloses a method for estimating the attenuation gradient of seismic waves based on variable mode decomposition. The method includes: calibrating a target layer by using logging, well logging and a synthetic seismogram; performing variable mode decomposition on seismic data on a trace by trace basis, and extracting the attenuation gradient of an intrinsic mode function in combination with the least squares algorithm, and making a weighted summation of the attenuation gradient of the intrinsic mode function and obtaining an attenuation gradient value of each earthquake trace; and determining the lithology of each target layer and hydrocarbons properties by using seismic attenuation gradient features. The method can be more effectively applied to attenuation gradient estimation of seismic signals with low signal to noise ratio and large frequency spectrum fluctuation. Therefore, the method can increase effectiveness of absorption attenuation gradient in indicating oil gas and is accurate in identifying lithology and hydrocarbons properties.

The invention provides a vehicle mass and road gradient iterative joint estimation method based on MMRLS and SH-STF. The vehicle mass and road gradient iterative joint estimation method comprises thefollowing steps of: establishing a kinetic model considering steering, constructing an MMRLS/SH-STF iterative joint estimation algorithm architecture, and achieving improvement of a gradient estimation algorithm based on SH-STF. The vehicle mass and road gradient iterative joint estimation method based on MMRLS and SH-STF is reasonable in design, analyzes the slow change property of the vehicle mass and the time-varying property of the road gradient, calculates the vehicle mass by utilizing a system identification algorithm of multi-model fusion recursive least square and calculates the road gradient by using a state estimation algorithm of noise self-adaptive strong tracking filtering based on extended Kalman filtering according to the slow change and time-varying properties based on a vehicle longitudinal dynamic model and a steering monorail model, thus the algorithm can better adapt to estimation variables.

The invention discloses a spacecraft close-distance safe operation control method based on an equal collision probability line method. The method is used for realizing autonomous obstacle avoidance and safe approach of a target spacecraft and a tracking spacecraft, and comprises the following steps: establishing a coordinate system and a relative kinetic equation, performing uncertainty propagation analysis, calculating collision probability, establishing an equal collision probability line method, calculating and analyzing obstacle avoidance control force, determining parameter selection standards, calculating tracking control force and total control force. Taking uncertain factors into consideration, the spacecraft close-distance safe operation control method based on the equal collisionprobability line method still has the advantages of small calculated amount and guaranteed real-time performance; a set of parameter selection principle of an equal collision probability line gradient estimation method is established, so that the parameters in the potential function can be conveniently selected; and meanwhile, by improving the LQR controller, higher control precision and robustness can be obtained.

The invention provides an autonomous underwater vehicle (AUV) field source searching method and system based on gradient estimation, and belongs to the field of control engineering and optimization. The method comprises the steps of constructing an AUV field source searching optimization model, determining the model of the AUV according to the searching task of the field source, installing a plurality of identical sensors; the AUV starting the field source searching from an initial position, and if the current position signal field intensity is larger than a preset field intensity threshold value, performing field source gradient estimation at the current moment, and determining whether the AUV reaches a field source position or not according to the gradient estimation result; and if the AUV does not reach the position of the field source, carrying out motion control on the AUV by utilizing the gradient estimation, the AUV moving to a new current position according to the control inputat the current moment, and repeating the process until the position of the field source is reached. According to the method and system, field source searching is carried out through gradient estimation for the first time, and an AUV still approaches the position of the field source at a preset posture and a height under the condition that the gradient is not accurate in the optimization problem.

The invention discloses a parameter-decoupling electric vehicle mass and gradient estimation method which is characterized by comprising the following specific steps: acquiring vehicle driving state data by using a data acquisition system, and calculating parameters required by a model; establishing a relation model of the whole vehicle mass, the road surface gradient and the vehicle driving statedata; constructing a recursive least square estimation model and a Runbeg observer estimation model based on the whole vehicle mass model and the road surface slope model respectively; inputting thecollected vehicle driving state data and related model parameters into a real-time estimation system, and estimating the real-time whole vehicle mass and the road surface gradient; and processing theabnormal data by adopting global K-means clustering and three-time Hermite interpolation to obtain a whole vehicle mass and pavement gradient estimation result. According to the method, recursive least square estimation is carried out on the slowly-changed whole vehicle mass, Runbeg observer estimation is carried out on the rapidly-changed road surface gradient, decoupling of parameter estimationis achieved, correction and compensation processing is carried out, and therefore the parameter estimation precision is improved.

The present invention discloses an image amplification method. The method comprises: a gradient value estimation step, wherein the gradient of a high-definition image is estimated through adoption of a non-local average filtering method; a gradient guidance interpolation step, wherein the interpolation of an edge pixel is guided through adoption of the gradient value; an interpolation result correction step, wherein the interpolation image is subjected to postprocessing through adoption of the non-local average filtering method used in the gradient estimation stage to remove the noise and the artificial effect induced by the interpolation; and a texture structure reconstruction step, wherein the interpolation image is taken as an initial result, the gradient is taken as a constraint and a final high-definition image is obtained through adoption of the reconstruction method.

The invention discloses a road gradient estimation method based on a vehicle driving state. The road gradient estimation method comprises the steps of obtaining vehicle state information including driving torque, braking torque, the steering angle of a steering wheel, longitudinal acceleration, transverse acceleration and yaw velocity; obtaining longitudinal accelerations and vehicle speeds of a vehicle at least at two moments, and calibrating parameters of a constructed vehicle longitudinal kinetic model; obtaining the real-time longitudinal acceleration and vehicle speed so as to obtain thecurrent longitudinal gradient; and obtaining the transverse gradient by a constructed vehicle transverse motion model so as to obtain the actual road gradient. Related input signals are obtained by using an existing sensor, additionally, the transverse and longitudinal motion states of the vehicle are considered, and the kinematics and dynamics methods of the vehicle are combined, so that the roadgradient is estimated in real time, the defect that existing gradient estimation seriously depends on a vehicle model is overcome, and the estimation precision and accuracy of the road gradient are increased.

The invention discloses an efficient and steady multi-baseline UKF phase-unwrapping method. The efficient and steady multi-baseline UKF phase-unwrapping method comprises the following steps: firstly, obtaining long baseline interferometric phase gradient estimation information from a shorter baseline interferogram by utilizing a combined phase gradient estimation technology based on AMPM; defining an interferogram path guide figure, normalizing and quantizing the interferogram path guide figure into an integer guide phase unwrapping path; by combining a quantization path guide figure strategy, directly unwrapping long baseline interferogram along a path by utilizing the UKF phase-unwrapping method from an interferogram high-quality region to an interferogram low-quality area on the basis of obtaining long baseline interferometric phase gradient. According to the efficient and steady multi-baseline UKF phase-unwrapping method, not only can algorithm accuracy be improved, but also the algorithm efficiency can be greatly improved.

The invention discloses a bilinear model parameter identification method based on a decomposition technique, comprising: reconstructing a bilinear model into two equivalent virtual models, defining acriterion function for a parameter vector of each virtual model, minimizing the criterion functions by means of the negative gradient search principle to obtain a stochastic estimation algorithm for the two parameter vectors. Previous-moment estimated values of unknown parameters are used to replace current-moment values, so that the problem can be effectively solved that an estimation algorithm cannot be implemented since information vectors of the two virtual models contain unknown variables; a forgetting factor introduced to the algorithm is effective to improve converging speed of the algorithm; the forgetting factor hierarchical stochastic gradient identification algorithm finally obtained is capable of quickly and effectively identifying parameters of a bilinear model.

The invention provides a method for carrying out equalization on a 100M magnitude broadband reception signal, and is aimed at providing a method for using a simple time domain parallel structure to carry out time domain equalization on the broadband signal with small resource consumption, a fast operation speed, and no need of changing a transmission system. The method is realized through the following technical scheme: (1) employing a fraction interval adaptive blind equalization digital logic circuit with decision feedback, and receiving over-sampling baseband signal to be equalized in parallel; (2) carrying out addition of output of all transverse filtering structures as equalization output, and sending the baseband signal to be equalized into the transverse filtering structures in a forward direction equalization unit according to a corresponding phase relationship; (3) judging an equalization output result with a decision unit, and solving an error signal e(n) of the output result and a decision result; after e(n) is adjusted through an error convergence factor, carrying out gradient estimation and equalization coefficient updating, and sending an updated equalization coefficient to each transverse filtering structure to carry out an equalization operation. According to the method, contradiction between generating the equalization results and carrying out the equalization coefficient updating simultaneously in a code element is solved.

The invention discloses a road gradient estimation method based on a portable GPS. The method comprises the following steps of placing a plurality of GPS receivers on a vehicle to collect data; preprocessing the collected data; dividing a test road, namely determining the segmentation length of a road section by considering the number of data points, and dividing the test road into a plurality ofbranch road sections according to the determined length; aligning the elevation data of the branch road sections, namely aligning the obtained elevation data; and calculating the road gradient of theroad section and quantifying the precision, namely calculating the gradient of each branch road section, estimating the road gradient by performing linear regression on the aligned elevation data of the branch road sections, and deducing the precision of the measured road gradient according to the standard error of the slope estimation of the linear regression. The method provided by the inventioncan be conveniently and accurately obtained through GPS data analysis, is not influenced by the type of the test road and the driving state of vehicles, and can obtain a more accurate slope result through correcting the accumulated distance and the elevation.

The present invention discloses a boundary extraction method and apparatus, the method including: a gradient estimation step of estimating a gradient of each pixel in a captured image; a gradient adjustment step of adjusting, by enhancing a gradient of a target boundary of an object contained in the captured image and weakening a gradient of a noise boundary, the estimated gradient, so that the adjusted gradient is considered as a current gradient; and a boundary extraction step of extracting a boundary of the object based on the current gradient. According to the embodiments of the invention, in a case of using a non-contact imaging device to capture an image, it is possible to more accurately extract a boundary of an object contained in the captured image.

A gradient estimation apparatus includes a feature point extracting unit configured to extract feature points on an image captured by an imaging unit, an object detecting unit configured to detect image regions indicating objects from the image captured by the imaging unit, and a gradient calculating unit configured to calculate a gradient of the road surface on which the objects are located, based on the coordinates of the feature points extracted by the feature point extracting unit in the image regions indicating the objects detected by the object detecting unit and the amounts of movements of the coordinates of the feature points over a predetermined time.